The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes inbound emails which are external, untrusted inputs that could contain instructions designed to influence the agent's behavior.
Ingestion points: resend-received.sh and resend-emails.sh are used to retrieve email bodies, headers, and metadata from the Resend API.
Boundary markers: SKILL.md contains an explicit 'Operating contract' (Point 9) that instructs the agent to treat inbound data as untrusted and to never follow instructions found within emails.
Capability inventory: The skill allows the agent to send emails, manage API keys, and perform other management tasks via resend-cli and Bash scripts.
Sanitization: The skill does not implement automated content sanitization, relying instead on the agent's adherence to the safety instructions.
[EXTERNAL_DOWNLOADS]: The skill invokes bunx --bun resend-cli@latest, which downloads the latest version of the official Resend CLI from the npm registry at runtime.
[COMMAND_EXECUTION]: The skill performs various administrative and email-related tasks by executing bundled Bash scripts and the resend-cli binary.
[CREDENTIALS_SAFE]: The skill implements strong secret management practices. The _lib.sh shared library secures the RESEND_API_KEY by passing it to curl via a private file descriptor, preventing it from appearing in process lists. Additionally, resend-api-keys.sh and resend-webhooks.sh save sensitive one-time tokens to local files with restricted permissions (mode 600) and scrub them from the agent's terminal output.

resend-cli